The basics of scikit-query

This page explains how to install the library and use it in Python.

Installation

The library can be installed using pip :

pip install scikit-query

Using the library

The library has modules for each kind of constraint (pairwise or triplet), plus another one containing the oracles. They can be imported as below :

from skquery.pairwise import *
from skquery.triplet import *
from skquery.oracle import MLCLOracle

This will allow to use the constraint selection algorithms as well as the MLCLOracle to answer queries about pairwise constraints.

Making queries

All algorithms have a fit method taking as arguments a matrix of n points having m features and an oracle (typically from the skquery.oracle module). The oracle must have a query method returning a boolean.

qs = AIPC()
oracle = MLCLOracle()
constraints = qs.fit(dataset.data, oracle)

The oracle’s truth attribute can support a ground truth labeling of the data, which will be used to automatically answer queries. If none is provided, it will ask queries to the user through the CLI.

oracle = MLCLOracle(truth=labels)

The constraints are returned as a dictionary of constraint types paired with lists of selected constraints. The table below describes how the constraint dictionary is structured.

Conventions used for constraint storage

Type	Key	Constraint format
Must-link	ml	(int, int)
Cannot-link	cl	(int, int)
Triplet	triplet	(int, int, int)